Geo Week: Clear SKAI and its Waves
By Madison Koehler
Severe natural disasters such as hurricanes, tornadoes, floods, and wildfires are not going away. In fact, climate experts expect these events to become more frequent and more severe due to climate change. Getting ahead of these disasters is crucial to protecting homes, businesses, and lives. New technologies designed to prepare for these disasters and respond quickly are a major topic of discussion within the geospatial field. As a presenter and panelist at Geo Week 2024 in Denver, I engaged in this discussion, highlighting how disaster response utilizing AI applications could make a real difference. And there appears to be an appetite for them.
Geo Week: The Intersection of Geospatial + the Built World
Geo Week is described as the premier event for increased integration between the built environment, advanced airborne/terrestrial technologies, and commercial 3D technologies. This conference is at the center of new technological innovations, hardware breakthroughs, and remote workflows that redefine expectations across teams, organizations, and entire industries.
My goal for the conference was to learn and share ways to integrate analytical tools and machine-learning technologies into the geospatial industry. My presentation, Emerging Technologies for Disaster Response, showcased how AI and other advanced analytical techniques have a critical role in disaster response, emphasizing that actionable AI insights in geospatial problems are accessible via tools such as our Clear SKAI app.
The Star of the Show — Clear SKAI
There are two main functions of the Clear SKAI App for disaster response: 1) When disaster strikes, timely damage assessment. 2) Availability of streamlined reporting of where damaged buildings and structures are.
Clear SKAI is based on the AI framework, SKAI. SKAI was developed by Google Research in partnership with the World Food Program to streamline the manual and time-consuming process of identifying damaged buildings and infrastructure to facilitate the fast rescue of individuals and efficient deployment of humanitarian resources. Clear SKAI is the application that surfaces what has been damaged and creates reports for first responders and disaster management personnel.
Timely damage assessment is directly linked to emergency responders’ ability to make decisions and take action quickly to deploy search-and-rescue operations and distribute resources, like fresh water and food, to those affected. However, manual damage assessments often take up to 3–5 weeks, whereas emergency responders need these decisions ideally within 48 hours of the disaster occurring. This is where SKAI and Clear SKAI come into play.
SKAI uses AI to automatically identify geospatial locations of buildings and infrastructure, something that humans have had to do manually. Then, it applies a change-detection model to flag the outlined buildings and infrastructure that are damaged through a comparison of before-disaster and after-disaster images. Aerial and/or satellite RGB imagery of a high resolution are the data that make SKAI scalable. Clear SKAI sits on top of SKAI and is the interface that produces the assessment reports.
The Google Research team demonstrated the utility of these imagery types for earthquakes, hurricanes, and wildfires. We leveraged the research and validated that the SKAI framework produces accurate tornado assessments.
We’ve taken the SKAI framework and made it more consumable through the Clear SKAI tool. Clear SKAI provides an interface with interactive maps that response teams can access with granular and accurate AI-driven assessments of damage. Clear SKAI’s maps and reports enable fast distribution of resources to regions using disaster-response criteria. The maps provide insights into the overall scope of the disaster, locations of damage hot spots, and damage to any particular buildings of interest like hospitals or schools.
Current computing technology makes AI timely and attainable. The application and tailoring of this tool to real-time situations requiring building-damage assessments is made possible by Google Earth Engine’s image collection and processing power, the availability of high-resolution post-disaster imagery (from sources like Planetscope), and Google Vertex AI’s model training pipeline.
Case Studies
The SKAI framework was first used in a real-time natural disaster in September 2022. Hurricane Ian, where catastrophic damage affected the west coast of Florida, was the use case. Due to SKAI’s timely damage assessments performed by Google Research and World Food Program, GiveDirectly was able to distribute relief funds to over 3,000 low-income households within one week of the hurricane making landfall. SKAI’s quick assessment enabled GiveDirectly to release these funds faster and target the funds to the most severely damaged neighborhoods.
We used the Clear SKAI tool to do a case study based on the 2020 tornado outbreak that affected several portions of Tennessee, focusing on Cookeville. The analysis area covered approximately 45 miles, including 263 damaged buildings and more than 6,000 pre-disaster and post-disaster aerial images. The imagery from before and after the tornado outbreak was compared to detect changes in the images of the same buildings, signaling damage.
Drill-down analytics were performed based on census-population tracts. Disaster response teams commonly use the tracts to guide effective emergency response and recovery operations. The highlighted details in Figure 3 include information about the population, the total number of buildings present, and the total number of both damaged and undamaged buildings.
Through these case studies, we can see how AI plays a critical role in disaster response, and machine-learning-driven decisions help enable timely responses to save lives and optimize resource distribution.
Clear SKAI Shakes Up the Panel
Following my presentation on Clear SKAI, I spoke on a panel alongside two other presenters who work on technologies for timely disaster response. The great engagement from the audience was a highlight of my trip!
With a large focus across the panel on the importance of acquiring timely data to extract speedy insights, Clear SKAI stood out for its access to a large collection of imagery through Google Earth Engine and other resources for post-disaster imagery. Clear SKAI also resonated with the audience for its ability to be quickly tailored to real-time situations needing building damage assessments via the time-efficient training of a new model specific to each situation to achieve the most accurate results possible.
It was gratifying to see a high level of enthusiasm from the audience about machine learning and its applications to these technologies, as well as excitement for what the future holds in this space.
Conclusion & Next Steps
A next step for Clear SKAI is to integrate disaster victims’ social media feeds into disaster-response efforts. We live in a connected world where people reach out for aid through their mobile devices. Through natural-language processing of web-scraped feeds from platforms like X and Instagram, disaster-response personnel could get real-time information about victims in need, their location, and state. In general, this data would help save even more lives!
Along with ever-evolving AI and software tools, the physical technologies utilized to collect and process RGB images continue to change. The vehicles used to capture high-resolution aerial imagery can take days to collect and process the necessary data following a disaster. With an emphasis on timely data in the geospatial field, these technologies will continue to improve and change.
For instance, UAVs like drones are becoming more popular for capturing post-disaster aerial imagery due to the convenience of not needing a pilot. The increasing range, battery life, and camera technology associated with these UAVs are continuously improving, increasing their popularity. This makes it essential for the Clear SKAI tool to evolve and work hand-in-hand with the best physical technologies.
Other future feature enhancements to the Clear SKAI application include tailored drill-down metrics (such as those in Fig. 4) and reports. Other data sources can be rolled into reports, and additional geospatial features may be displayed. Through Object Computing’s Asterisms platform, the application is easily modified to fit users’ needs.
Exposing the geospatial community to practical applications of AI at the Emerging Technologies for Disaster Response panel discussion was an enriching experience. Geo Week was a jam-packed event that allowed me to relate my work with AI to the hot topics of the geospatial world and share my knowledge with other technology enthusiasts. Following this, I am more excited than ever to see what the future holds for the applications of AI in the geospatial world.
Learn More
Madison Koehler obtained her MS in Artificial Intelligence in 2022 and has spent the last year kicking off her career as a data scientist. She utilizes a strong background in mathematics, statistics, and computer science to create data-driven insights in the machine learning and deep learning space, and has utilized cloud service providers such as Amazon Web Services (AWS) to build and optimize large-scale pipelines spanning the entire machine learning lifecycle.
